Runbook

# A Design Sketch Commands: - run runbook X - run runbook template X that is instantiated with given inputs - show an execution plan for runbook X (dry-run) - convert a conversation into a runbook # Rest API Specification - `/v1/organizations` - `/v1/organizations/<org-id>/projects` - `/v1/organizations/<org-id>/projects/<project-id>/runbooks` - `/v1/organizations/<org-id>/projects/<project-id>/runbooks/<runbook-id>/logs` # MCP Server Specification ## Resources Runbook: - Path: `runbook://<host>/<runbook-id>` - Type: text resources Runbook execution log: - Path: `runbook-execution-log://<runbook-id>/logs/<log-id>` - Type: text resources ## Prompts For each runbook, a corresponding prompt is created. Here is an example request. ``` { method: "prompts/get", params: { name: "execute-runbook", arguments: { runbookName: "backend-service-deploy" } } } ``` ## Tools - `create_runbook` ## Roots TBD # Alternative Design Considered Instead of using Prompts, we initially considered making the Runbook MCP server act as a MCP client so that it can execute a runbook by interacting with other MCP servers. Sampling is another approach that allows the MCP server to talk to LLM, but it is not currently supported in the Claude Desktop client. We might revisit the design, but for now, we will see if Prompts are sufficient to handle our use cases. # Development Notes - All the actions taken are stored in an audit log. This requires some more thoughts. - We would like to support the dry-run mode if possible. https://github.com/modelcontextprotocol/specification/issues/97 might be related. - We would like to recursively call MCP if possible. https://github.com/modelcontextprotocol/specification/discussions/94 is somewhat relevant. See also https://modelcontextprotocol.io/development/roadmap#agent-support ## Virtual Env ``` uv venv source .venv/bin/activate uv add "mcp[cli]" httpx ``` ## Inspecting the content of the databs